AMD RX 7000 series GPU Owners' Club

[kapone32]

Does anybody here have a 7600 (XT) here? If so, how does it do with video playback power consumption? Is it as bad as chiplet-based RDNA 3 cards?

I'm thinking about upgrading my spare 6500 XT to a 7600 because it has a better video decoder, better performance, and I could do it with minimal investment, but there's no point if it eats up to 50 W playing a video like the rest of the 7000-series.

For power efficiency you are not going to see anything like the 6500XT. It won't be 50 Watts but it will still be more than double what the 6500XT outputs.

 

[AusWolf]

For power efficiency you are not going to see anything like the 6500XT. It won't be 50 Watts but it will still be more than double what the 6500XT outputs.

Oh, that's a shame. So it seems it's an architectural problem within RDNA 3, not one originating from the chiplet design, right?

 

[Vya Domus]

So it seems it's an architectural problem within RDNA 3, not one originating from the chiplet design

RX 7600 has a chip twice as larger as the one the 6500 XT uses, your expectations are very out of line, of course it wont have the same power draw.

 

[AusWolf]

RX 7600 has a chip twice as larger as the one the 6500 XT uses, your expectations are very out of line, of course it wont have the same power draw.

Of course it has twice as many shaders, but I would expect both chips to use their video decode engines, not the shader cores for video playback. Or is this not how it works? (Please excuse my lack of knowledge of the technicalities of video playback on GPU.)

 

[Vya Domus]

Of course it has twice as many shaders, but I would expect both chips to use their video decode engines, not the shader cores for video playback. Or is this not how it works

Regardless a chip with more than twice as many transistors on similar nodes will use more power, especially in idle, it's unavoidable.

 

[Super Firm Tofu]

Of course it has twice as many shaders, but I would expect both chips to use their video decode engines, not the shader cores for video playback. Or is this not how it works? (Please excuse my lack of knowledge of the technicalities of video playback on GPU.)

It’s the architecture. The 7600 uses as much power for video playback as a 4090. The number of transistors has nothing to do with it.

 

[AusWolf]

It’s the architecture. The 7600 uses as much power for video playback as a 4090. The number of transistors has nothing to do with it.
View attachment 355120

That's a shame. I would have expected newer drivers to improve the situation since the TPU review.

 

[Super Firm Tofu]

That's a shame. I would have expected newer drivers to improve the situation since the TPU review.

I’m not sure what the 6500XT draws, but 23w isn’t that bad. Seems like a fair trade for better overall performance.

 

[AusWolf]

I’m not sure what the 6500XT draws, but 23w isn’t that bad. Seems like a fair trade for better overall performance.

I'm not sure considering that RDNA 2 only measures GPU chip power consumption and I don't have the tools to measure it across the whole board. It's the VRAM that seems to be running at a higher frequency during video playback.

The overall performance is a good point. Let's see if I can sell the 6500 XT for a fair price.

 

[Vya Domus]

The number of transistors has nothing to do with it.

Yes it absolutely has everything to do with that lol

7600XT - Die Size : 204 mm², Transistors : 13,300 million
6500XT - Die Size : 107 mm² Transistors : 5,400 million

It's not the architecture, you're comparing GPUs where one is twice as large, gee I wonder which one will use more power.

 

[kapone32]

Yes it absolutely has everything to do with that lol

7600XT - Die Size : 204 mm², Transistors : 13,300 million
6500XT - Die Size : 107 mm² Transistors : 5,400 million

It's not the architecture, you're comparing GPUs where one is twice as large, gee I wonder which one will use more power.

There is also the fact that the 6500XT idles at 2 Watts in idle.

 

[Vya Domus]

7600 got twice the amount of VRAM as well, even if the chips were to be identical I'd still use more power for obvious reasons.

 

[Super Firm Tofu]

Yes it absolutely has everything to do with that lol

7600XT - Die Size : 204 mm², Transistors : 13,300 million
6500XT - Die Size : 107 mm² Transistors : 5,400 million

It's not the architecture, you're comparing GPUs where one is twice as large, gee I wonder which one will use more power.

The issue with high power draw for video playback is an architecture trait of RDNA3. @AusWolf was asking if it was caused by the chiplet design alone. It's not. It's RDNA3 specific.

Look at the 6700XT - 335 mm² Die Size, 17,200 million transistors, TSMC 7nm, 12GB DDR6, and video playback power is 15W.

 

[Kairakhan]

I live in Ankara, Turkey. I bought RX 6600 two years ago and switched to RX 6700 XT in October 2023. I upgraded my Msi pro m650 to wifi motherboard, Ryzen 5 7600 processor, 32 GB ddr5 crucial 5600 mhz ram, adata pylon 750 watt system to Sapphire Rx 7900 xt nitro.

 

[Vya Domus]

was asking if it was caused by the chiplet design alone. It's not.

He was asking about 6500XT vs 7600 specifically, there is no way the 7600 was ever going to draw as much or less power than a 6500XT due to their very different size, TDPs, and VRAM capacity, those are the very obvious reasons, the architecture would be by far the last on that list, this isn't up to debate.

Btw the 7600 doesn't even have chiplets. We'd need cards with the same VRAM and chip size to conclusively determine how much of this comes down to the architecture.

Just look in the graph you posted and stop arguing about this, 40 series cards that have larger chips and more memory consume more power in video playback and they obviously have the same architecture.

 

[AusWolf]

He was asking about 6500XT vs 7600 specifically, there is no way the 7600 was ever going to draw as much or less power than a 6500XT due to their very different size, TDPs, and VRAM capacity, those are the very obvious reasons, the architecture would be by far the last on that list, this isn't up to debate.

Btw the 7600 doesn't even have chiplets. We'd need cards with the same VRAM and chip size to conclusively determine how much of this comes down to the architecture.

Just look in the graph you posted and stop arguing about this, 40 series cards that have larger chips and more memory consume more power in video playback and they obviously have the same architecture.

Actually, my question was a bit of both. I never expected the 7600 to use less power than the 6500 XT even in video playback, but I was curious if it was at least somewhat comparable due to being a monolithic chip, or if it wasn't comparable due to the architectural differences.

You see in the linked chart that even the 4090 uses less power in video playback than the 7600, so it's not only a matter of GPU size and the number of RAM chips. Modern GPUs have video decoders built-in, so I don't see why they should use their shader cores and ramp up clocks and power to play video, but apparently, RDNA 3 does just that. It's very strange.

Edit: I was also curious if newer drivers alleviated the matter a little since the TPU review.

 

[wNotyarD]

Actually, my question was a bit of both. I never expected the 7600 to use less power than the 6500 XT even in video playback, but I was curious if it was at least somewhat comparable due to being a monolithic chip, or if it wasn't comparable due to the architectural differences.

You see in the linked chart that even the 4090 uses less power in video playback than the 7600, so it's not only a matter of GPU size and the number of RAM chips. Modern GPUs have video decoders built-in, so I don't see why they should use their shader cores and ramp up clocks and power to play video, but apparently, RDNA 3 does just that. It's very strange.

Edit: I was also curious if newer drivers alleviated the matter a little since the TPU review.

Issue with any Radeon isn't shader clock. It's VRAM clock gating.
AMD simply can't get VRAM to clock lowly without weird stuff happening, so any activity as light as it may be makes the memory clock shoot up.

 

[AusWolf]

Issue with any Radeon isn't shader clock. It's VRAM clock gating.
AMD simply can't get VRAM to clock lowly without weird stuff happening, so any activity as light as it may be makes the memory clock shoot up.

I see, but why is it so much worse on RDNA 3 compared to previous generations? I had my doubts if it was due to the chiplet design, but it's apparently not, as the 7600 is equally affected.

 

[b0uncyfr0]

Where do folks in the EU get reliable PTM7950 from?

I want to reduce my hotspot by even 5 degree's and im out of options. Already undervolted, reduced power which is the worst culprit for my nitro 7800xt/case.

Ive got some NT-H2 i might try first though. I had a difficult time getting the shroud off the nitro 7800xt . They're not standard screws, what are they?

 

[AusWolf]

Where do folks in the EU get reliable PTM7950 from?

I want to reduce my hotspot by even 5 degree's and im out of options. Already undervolted, reduced power which is the worst culprit for my nitro 7800xt/case.

Ive got some NT-H2 i might try first though. I had a difficult time getting the shroud off the nitro 7800xt . They're not standard screws, what are they?

What's your hotspot now? If it's below 100 (or even 90) degrees, I wouldn't bother. You could do more harm than good. Or just get some Kryosheet.

Sorry for not answering your question.

 

[b0uncyfr0]

What's your hotspot now? If it's below 100 (or even 90) degrees, I wouldn't bother. You could do more harm than good. Or just get some Kryosheet.

Sorry for not answering your question.

About 85-87 ish with the backplate off + wrong orientation.

Under 75 with proper orientation.

 

[AusWolf]

About 85-87 ish with the backplate off + wrong orientation.

Under 75 with proper orientation.

That's more than fine. Don't worry about repasting.

 

[b0uncyfr0]

That's more than fine. Don't worry about repasting.

I spoke too soon - the hotspot is now hitting 96-97.

I really wanna keep it under 90 with my current OC.

 

[AusWolf]

I spoke too soon - the hotspot is now hitting 96-97.

Is that with OC? If so, it's still not bad.

I really wanna keep it under 90 with my current OC.

How hot does it get at stock? Under 90 with a heavy OC sounds to me like a pretty harsh expectation.

 

[b0uncyfr0]

Is that with OC? If so, it's still not bad.


How hot does it get at stock? Under 90 with a heavy OC sounds to me like a pretty harsh expectation.

Im running 2700 core at 15% power / 1030mv. Fast memory timings at 2514.